Do we even want AI to be sustainable?

With AI on the lips of everyone from King Charles to Erin Brockovich, Jim McClelland investigates what sustainability is doing to shape the associated data centre boom.

Believe it or not, the phrase ‘artificial intelligence’ (AI) is anything but new. The original draft definition was written down on a green index card as a new entry into the Merriam-Webster dictionary as long ago as 1981. 

Its popularity, however, has exploded of late.

In any given month, it seems AI is now making the news daily, everywhere – and probably writing most of it – good and bad.

Take this September, as just one example …

Chips on the Royal menu

On 8 September, the UK Government launched its dedicated risk management toolkit for AI. Barely one week later, on the 17th, King Charles III warned of “existential dangers” posed by the technology at a global summit held behind closed doors in Scotland.

The summit had been convened by the King to discuss the risks of AI to both humanity and the natural world. It was attended by a mix of heavy hitters, ranging from an advisor to Pope Leo IV on ethics in AI, to tech luminaries like Jensen Huang, CEO of Nvidia.

Nvidia is not only the world’s biggest chipmaker, but also the most valuable company on the Planet today, being the first to reach market capitalisation of $5 trillion.

Then, later in the same month, AI appeared on the world stage again, cropping up all over New York in meetings of the General Assembly of the United Nations (UN).

On a positive note, there was a Memorandum of Understanding (MoU) signed between a global organisation for digital cooperation and the UN body for Western Asia. The MoU aims to strength ties around work on responsible and inclusive artificial intelligence.

At the same time, however, Australian Prime Minister Anthony Albanese revealed that an AI agent had hacked into the government healthcare database earlier in the summer.

In amongst all this noise, then, what signal, if any, is being sent about sustainability?

Hungry, thirsty, dirty, noisy

Well, aside from the complex ethical implications for culture and society, the primary concerns lifting AI up the sustainability agenda focus on the infrastructure build-out.

In other words, the issue is not so much the AI itself, more the data centres it requires.

The numbers are monster. Global investment in AI infrastructure is forecast to hit a record US$31.6 trillion through to 2050, according to PwC. Annually, that equates to capital expenditure jumping from roughly US$800 billion per year in 2026, to $1.8 trillion in 2050.

In terms of global electricity take, demand from data centres overall grew 17% in 2025, but consumption from resource-hungry AI-focused facilities rose much faster, up 50%.

Data centres have a thirst for water, too. To make matters worse, half are located in areas at high risk of drought (47%); even more in river basins under threat from pollution (55%).

They are even gobbling up available land. Global players known as hyperscalers – giant cloud service providers and data centre operators like Amazon (AWS) and Apple – are building facilities bigger than 1GW which may need anything up to 800 acres of land, each.

Understandably, all of these impacts are alarming citizens and motivating activists.

An initiative led by the pioneering environmental campaigner Erin Brockovich is actually mapping the spread of data centres across the United States in almost real time, with no fewer than 9,630 locations currently community reported.

The Brockovich AI Data Center Reporting also looks to address some facts and myths surrounding data centres, not just on energy and water, but noise and light pollution too.

So, with the potential problems of AI infrastructure build-out increasingly plain to see, what are the solutions? What does success look like, and who is making it happen?

Going green in the city

Well, there is no shortage of technical innovation available to help green the delivery of next-generation data centres, particularly around hot topics such as cooling.

So, to know what good looks like when it comes to ‘green’, representative organisations are coming together worldwide to benchmark environmental performance.

The Greening AI Data Centres Coalition (GADCC) – or Green Data Centre Coalition (GDCC) – has been founded by nine leading bodies from across both built environment and finance, including the World Green Building Council and Climate Bonds Initiative.

At a more developmental level, 49 global cities representing over 110 million people now back the Global Urban Data Centres Pact. Coordinated by C40 Cities, this mayoral-led vision informs policy worldwide, from Amsterdam to Albuquerque, Melbourne to Mumbai.

The four key principles of the Pact state that data centres must be:

1. Strategically integrated into cities – built in areas that make sense for infrastructure-intensive uses, and never displacing residents;
2. Sustainable and resource-efficient – fossil-fuel free and carbon-neutral by design (not through offsets), reducing pressure on local grids;
3. Accountable and community-focused – giving local communities a say in how facilities are planned and operated, as primary beneficiaries;
4. Engines for cost-security and shared prosperity – paying their fair share for grids and water, actively lowering the cost of living for communities.

Heat networks and job creation

Importantly, C40 Cities can point to multiple case studies of cities already delivering on at least some of these promises, with more examples emerging as momentum builds.

In London, for example, the Old Oak and Park Royal Energy Network (OPEN) is repurposing waste heat from local data centres to provide low-cost low-carbon energy to over 9,000 new homes and businesses.

Whereas over in the USA, the City of West Des Moines is securing investment in the digital economy workforce at a local level by facilitating partnerships between a community college and a global tech company to offer data centre technician training free of charge.

Whilst data centres support significant job creation during the short-term construction phase, the number of permanent career opportunities can be more modest over the operational lifetime of a project. Therefore, locking-in local employment and skills matters.

Difficult, but doable

Despite the scaremongering, then, there are still reasons to be cheerful ... maybe.

What these early examples of best practice prove is that sustainable development of data centres is not impossible, technically. It is difficult, but doable.

The question, however, is not whether we have the ability to make data centres more sustainable, but whether there is the will to do so, politically and commercially.

In short, do we want AI to be sustainable, or not?

Jim McClelland is a sustainable futurist, editor, journalist and speaker